A pytorch implementation of the paper ''UNI-IQA: A Unified Approach for Mutual Promotion of Natural and Screen Content Image Quality Assessment"

Python 3+
PyTorch 1.4+
Matlab

data_all.m

combine_train.m

python main.py

Compute SRCC/PLCC after nonlinear mapping: result_analysis.m
Compute fidelity loss: eval_fidelity.m

python demo.py

We utilize both NI and SCI datasets in the experiment.
NI datasets: LIVE, CSIQ, KADID-10K, TID2013, LIVE-Challenge and KonIQ- 10K
SCI datasets: SIQAD and SCID
(We will present links to download these datasets for easy access.)

Our code is base on UNIQUE, we are truly grateful for the authors' contribution.